New assessment finds Oxford-developed technology could accelerate diagnosis and improve survival
An Oxford-developed AI tool could identify a widely underdiagnosed form of heart failure almost nine months earlier than standard care, according to an assessment. The findings come from the first report by the American Heart Association’s AI Assessment Lab, which examines whether clinical AI can improve patient outcomes and deliver financial value rather than focusing solely on technical accuracy.
The report assesses EchoGo Heart Failure, an AI-powered diagnostic support tool developed by Oxford-based Ultromics. Drawing on approximately 90,000 real-world echocardiograms supplied by Dandelion Health, the analysis models how the technology could improve the diagnosis and management of heart failure with preserved ejection fraction, known as HFpEF.
The findings suggest EchoGo Heart Failure could identify the condition an average of 263 days earlier among patients who would otherwise experience a delayed diagnosis. Over five years, earlier intervention could prevent 477 deaths for every 10,000 patients, while reducing hospital admissions and emergency department visits.
Identifying an underdiagnosed condition
HFpEF occurs when the heart contracts normally but does not relax sufficiently to fill with blood. Patients may experience breathlessness or fatigue, but the symptoms can resemble those of other conditions and may not be detected through standard imaging.
Diagnosis is particularly difficult because patients generally retain a normal or near-normal ejection fraction, which measures the proportion of blood pumped from the heart’s main chamber during each contraction. As a result, HFpEF may remain undiagnosed until symptoms have progressed.
The condition is also associated with inequalities in diagnosis. Normal ranges used in cardiac assessment have historically been derived predominantly from white male populations, potentially contributing to underdiagnosis among women and people from minority ethnic backgrounds. An estimated 30 to 50 per cent of patients are unaware they have heart failure, according to the press release accompanying the report.
EchoGo Heart Failure analyses a single apical four-chamber video clip taken during a standard transthoracic echocardiogram. The cloud-based system detects patterns associated with HFpEF and produces a classification that clinicians can consider alongside other patient information.
The platform works with existing echocardiography equipment and hospital imaging systems, without requiring additional hardware. Ultromics was founded at the University of Oxford, while the technology was developed in partnership with the NHS and Mayo Clinic. EchoGo Heart Failure received clearance from the US Food and Drug Administration in 2022.
Earlier diagnosis could improve outcomes
The American Heart Association analysis compared patients whose HFpEF was identified shortly after an echocardiogram with those whose condition was initially missed but subsequently detected by the algorithm. Patients whose diagnosis was delayed could potentially have been identified 263 days earlier if EchoGo Heart Failure had been used at the time of their original scan.
Earlier recognition could allow appropriate treatment to begin sooner and reduce the risk of deterioration. For every 10,000 patients over five years, the model projected 406 fewer hospital admissions and 501 fewer readmissions. It also estimated 564 fewer emergency department visits.
Subgroup analysis suggested the technology could offer greater benefits for younger and non-white patients. More consistent interpretation of routine cardiac imaging may therefore help reduce some of the disparities affecting groups whose symptoms have historically been overlooked.
The report also examined the financial implications for providers and payers. Savings were estimated at approximately $1,800 per patient, while a representative health system could generate up to $1.9m in additional revenue over five years.
The financial modelling included reimbursement for using the algorithm and the revenue associated with earlier diagnosis and management. According to the report, the resulting income could outweigh revenue reductions associated with lower use of acute hospital services.
Building confidence in clinical AI
The AI Assessment Lab has been established to give healthcare providers clearer information about the potential value of AI technologies in routine care. Although regulatory clearance can establish safety and performance, it does not necessarily show how a product will affect clinical pathways or health system finances.
“The American Heart Association AI Assessment Lab provides an independent evaluation and testing environment for clinical AI algorithm performance as well as economic impact, patient outcomes and healthcare workflows,” said Dr Lee H Schwamm, a stroke neurologist and member of the association’s AI Solutions in Healthcare Steering Committee.
The lab is guided by an external advisory panel that includes specialists in AI and cardiovascular medicine, alongside experts in healthcare delivery. Its assessments are intended to help hospitals determine which products can support their clinical and financial objectives.
Dandelion Health provides the real-world clinical and imaging data used in the assessment. The company’s Chief Data Officer, Dr Shivaani Prakash, said large-scale data was essential to understanding how diagnostic technologies perform in practice.
“When algorithms are rigorously and independently tested on diverse, real-world data, and the results are easy for providers to understand, AI developers are incentivised to build better tools, and providers are empowered to adopt the ones that will deliver the most value for their patients,” she said.
From projected benefits to clinical practice
The findings are projections based on retrospective data and health economic modelling rather than outcomes from a prospective clinical trial. Results may vary according to local populations, care pathways and whether patients receive further assessment and appropriate treatment.
The assessment does not verify the algorithm’s accuracy or compliance with clinical and regulatory standards. American Heart Association Ventures has previously invested in Ultromics, although the AI Assessment Lab operates independently using a standardised methodology.
Despite these limitations, the report shows how evaluating potential patient outcomes and financial impact could help providers identify AI technologies capable of delivering meaningful improvements in care.
Have some thoughts on the topic? Share them with the community here